Spiral Calendar Clusters No Better Than Chance.

Posted: April 1st, 2012 under Stocks.

I hesitate to post this on April 1st, because I expect people will take it as a joke, however I’ve recently come to the conclusion that spiral calendar clusters are no more likely to match market highs or lows than dates chosen at random. As evidence I have added some statistics to the chart showing the percentage of spiral clusters that match a high or low, either to the day, or on the second line, within 1 day plus or minus. I also show for any date within the window chosen at random, the chance of that date being a high or low. For the approximately 24 year period shown, spiral clusters in the S&P 500 are almost exactly as good as dates chosen at random at predicting highs and lows. Any visual evidence of frequent matches, such as I posted in the Citigroup chart can be attributed to chance, since smaller samples may appear to do better without being statistically significant.

The number of spiral calendar clusters we would expect to match highs or lows probably should be treated as a Bernoulli process statistically, although I am not certain about the assumption of independent trials. The chance number quoted in the chart is just the chance of a single trial, or single date chosen at random, being a high or low.

Spiral Calendar Alignment Hit Probability For S&P 500, 1988-2012

Spiral Calendar Alignment Hit Probability For S&P 500, 1988-2012 – click chart to enlarge.

The chart may be distributed according to the Creative Commons by attribution commercial license 3.0 provided the Spiraldates.com legend is retained in its original form and/or the entire chart is a link to Spiraldates.com.

Be Sociable, Share!


  1. Congratulation for your job of this months!It was very helpful for me and very often it takes important change in trend.Haw can I have the map of April?

    Comment by paolo — April 4, 2012 @ 5:12 am

  2. Surely with the evidence of randomness one should be able discern probability. The aim of the forescast is not be 100% correct, but to identify the odds. I cant believe that you would say after publishing the a forecast 30 days in advance, have evidence to show that your forecast where correct more than 50% of the time and then say the forecasting does not work. I have looked at the date clusters that were published over the last 4 months and its more accurate than what I have read about on published online media. Is there another angle from which one can explain why this does not work?

    Comment by Shane — April 6, 2012 @ 9:47 pm

  3. I’m still working on the stats to get a better number for the chance of getting X hits in some interval instead of just using the chance of a single prediction being a hit.
    Also slicing/dicing different ways with different variables and other filters to see if there might be a subset that works better. However at this point I unfortunately don’t have anything to refute the broad conclusion I made in the post.


    Comment by Admin — April 6, 2012 @ 10:09 pm

RSS feed for comments on this post. TrackBack URL

Leave a comment